nf7 v2.0 hypertransport: unidirectional or bidirectional?

Subject says it all. Do Nforce2 boards feature a unidirectional or
bidirectional hypertransport interface?

--
Doug

pigdos wrote:
> Subject says it all. Do Nforce2 boards feature a unidirectional or
> bidirectional hypertransport interface?
>

I cannot find anything definitive. Datasheets are covered under NDA,
which means a person holding a datasheet, cannot discuss it with
you. The review sites only say the bus is "800MB/sec" without
elaborating whether that number is for one direction, both
directions, or whatever. I'd say Nvidia's secret is safe.

If you look at the layout, there appear to be sixteen diff pairs
on the top surface of the board. They are routed between the
Northbridge and Southbridge, and typically snake around the end
of the AGP connector, to get to the Southbridge.

Some pics here
http://www.tweak.dk/tests2.php?id=267

(16 diff pairs, middle right of picture, leading to upper
right hand corner, eventually leading to the Southbridge)
http://www.tweak.dk/test252/north.JPG

With 16 diff pairs, my *guess* would be two 8 bit busses.
But anything is possible. Certainly other product info
I've looked at (AMD8111), Hypertransport is two separate
busses, one transmit and one receive, and each consists
of diff pair signalling. Maybe if the speed dropped low
enough, a designer could consider going single-ended, but
then, what would be the point. Diff is so much better.

Paul

Paul, it's interesting to note that the Nforce1 boards had a 400MB/s
hypertransport interface (at least according to the specs I've read).

Doesn't the hypertransport spec call for a bidirectional interface?

--
Doug
"Paul" <> wrote in message
news:eigmsj$385$...
> pigdos wrote:
>
> I cannot find anything definitive. Datasheets are covered under NDA,
> which means a person holding a datasheet, cannot discuss it with
> you. The review sites only say the bus is "800MB/sec" without
> elaborating whether that number is for one direction, both
> directions, or whatever. I'd say Nvidia's secret is safe.
>
> If you look at the layout, there appear to be sixteen diff pairs
> on the top surface of the board. They are routed between the
> Northbridge and Southbridge, and typically snake around the end
> of the AGP connector, to get to the Southbridge.
>
> Some pics here
> http://www.tweak.dk/tests2.php?id=267
>
> (16 diff pairs, middle right of picture, leading to upper
> right hand corner, eventually leading to the Southbridge)
> http://www.tweak.dk/test252/north.JPG
>
> With 16 diff pairs, my *guess* would be two 8 bit busses.
> But anything is possible. Certainly other product info
> I've looked at (AMD8111), Hypertransport is two separate
> busses, one transmit and one receive, and each consists
> of diff pair signalling. Maybe if the speed dropped low
> enough, a designer could consider going single-ended, but
> then, what would be the point. Diff is so much better.
>
> Paul

pigdos wrote:
> Paul, it's interesting to note that the Nforce1 boards had a 400MB/s
> hypertransport interface (at least according to the specs I've read).
>
> Doesn't the hypertransport spec call for a bidirectional interface?
>

http://www.hypertransport.org/tech/tech_specs.cfm

Page 19 here, shows two unidirectional links. When operating
at extremely high speed, separate unidirectional links makes
good sense.

http://www.hypertransport.org/docs/s...OLink_Spec.pdf

Paul